Generative AI: A Powerful Tool for Advancing Oncology Clinical Practice

Corey Zankowski

Sep 23, 2024

4 min read

0

48

0

Photo by National Cancer Institute on Unsplash

As a technology enthusiast dedicated to improving access to cancer care, I can’t help but be excited about the potential of AI and Generative AI (GAI) to revolutionize cancer care. If you’ve been reading the popular press lately, you’ve read about OpenAI’s ChatGPT, GPT-3, GPT-4, Google’s Bard, or Meta’s LLaMA. I started reading about them in November 2022 and assumed it was another fad. When the articles didn’t stop, I thought it would be good to check out what everyone was talking about.

I started reading about how GAI is an AI system capable of creating new and unique content, such as text, music, and images, based on learning patterns from existing data. This creative power suggests that, in my field of healthcare, appropriately designed GAI tools can supercharge physicians. While I recognize we are currently in the middle of the hype cycle on GAI, my instinct is to dive head-first into this ocean of opportunity. At the same time, decades of experience with medical devices taught me to be cautious when patient lives are at stake, so we’d do well to put on a floatation device before jumping in. By next year, several commercially available healthcare solutions will incorporate GAI, probably introduced by smaller healthcare companies with less to lose. This article is the first in a series documenting the journey to learn about the responsible use of GAI in healthcare and uncover its promises and pitfalls as a tool to democratize cancer care so that Primum can be among the first movers.

AI has already shown impressive results in various domains of oncology, such as imaging, diagnosis, prognosis, treatment, and research. For example, radiology and therapy planning solutions already use AI to detect cancer from radiographic images, outline normal organs in CT and MRI images, and predict the optimal dose for radiotherapy treatment plans. Generative AI can generate synthetic images of tumors from different modalities, such as MRI or CT scans, where actual image data are hard to come by. GAI can also help classify and segment tumors based on histopathology slides or radiomics features and predict their molecular characteristics and clinical outcomes. Furthermore, GAI can assist in creating novel anticancer drugs by generating new molecular structures and optimizing their properties. GAI can also support clinical decision-making by generating personalized treatment recommendations based on patient data and preferences.

Large Language Models (LLMs), an important area of GAI, already show great potential to address the tsunami of medical literature published quarterly by summarizing complex, voluminous text. AI services that convert long articles into short summaries that retain essential and relevant information already exist. [1, 2, 3, 4] Given the rapid growth in oncology literature [5], oncologists feel overwhelmed with new information, and this feeling is acute in general oncologists who deal with various cancers daily. In this context, it is natural to apply GAI to transform how oncologists consume new information, making it easier to find what they need. Natural Language Generation (NLG) has the power to summarize and extract knowledge from complex biomedical literature, trials, and clinical content [1]. Recent advances in GAI have led to models capable of accurate and detailed text-based responses to written prompts (“chats”) [6, 7]. These models score highly on standardized medical examinations and have shown promise in clinical applications such as complex diagnostic reasoning. [6]

The big challenge for NLG will be interpreting a clinical question, summarizing existing text, and synthesizing a response without making things up (aka ‘hallucinating’). After all, NLG is ‘generative’ and is perfectly capable of making up semantically correct and very convincing gibberish. While gibberish might be helpful when writing science fiction, it can be life-threatening in a healthcare setting. As a cautionary tale, we’ve learned that using ChatGPT professionally without validating the generated content is a career-limiting move– just ask attorney Steven A. Schwartz who got disbarred for crafting a motion full of made-up case law. [7]

At Amplifai Oncology, we enable oncologists to collaboratively review the work of peers, with the help of AI. We are adopting Generative AI in our platform to assist oncologists and their care team in doing their jobs more cost-effectively. Starting with modest feature extraction and summarization and predicting the shape of an achievable dose volume histogram, to segmenting normal tissues and predicting a personalized 3D dose distribution for a specific patient, the possibilities are endless.

We will partner with leading AI researchers, top oncology specialists, and community oncologists to safely and effectively respond to their questions to achieve our goals. We will explore how harness its the potential of Generative AI while avoiding its pitfalls, and where it makes the most sense to deploy this technology to make the lives of oncologists and cancer patients better.

If you are interested in partnering with us on this journey, email info@amplifai-oncology.com.

References